private List<SimpleBicycle> getKnownModelList(String filename) {
    List<SimpleBicycle> list = new ArrayList<SimpleBicycle>();
    File file = new File(filename);
    try {
      InputStream inputStream = new FileInputStream(file);
      //            CsvReader csvReader = new CsvReader(inputStream, Charset.forName("CP1251"));
      CsvReader csvReader = new CsvReader(inputStream, Charset.forName("UTF-8"));
      csvReader.setDelimiter(';');
      while (csvReader.readRecord()) {
        String[] values = csvReader.getValues();
        SimpleBicycle simpleBicycle = new SimpleBicycle();
        simpleBicycle.setDirtyModel(values[0].trim());
        simpleBicycle.setModel(values[1].trim());
        simpleBicycle.setWheelSize(WheelSize.getSizeByValue(values[2].trim()));
        list.add(simpleBicycle);
      }

    } catch (FileNotFoundException ex) {
      ex.printStackTrace();
    } catch (Exception ex) {
      ex.printStackTrace();
    }

    return list;
  }
예제 #2
0
  /** Sets up all the GEPSymbolSet symbols, loading them from the parameter file. */
  public void setup(
      final EvolutionState state, final Parameter base, final Parameter def, GEPSpecies species) {
    // Name of file with the terminal (variable) definitions and training values
    String terminalFilename;
    // Name of file with the test data values if specified
    String testingTerminalFilename;

    // keep track of the maximum arity of any function
    maxArity = 0;

    // What's my name? Don't really use this at this time ...
    name = state.parameters.getString(base.push(P_NAME), def.push(P_NAME));
    if (name == null || name.equals(""))
      state.output.warning(
          "No name was given for this GEP symbol set...not required at this time.",
          base.push(P_NAME),
          def.push(P_NAME));

    // How many functions do I have?
    numberOfFunctions =
        state.parameters.getInt(base.push(P_FUNCTIONSIZE), def.push(P_FUNCTIONSIZE), 1);
    numberOfSymbols = numberOfFunctions;

    // How many terminals do I have? Check for a data file first ...
    // if time series problem type and using raw time series data then
    //      number of terminals will be specified in the embedding dimension value
    //      provided in the parameter file
    // else if a file specified
    //      get the 1st line of the file and count the fields in it (#terminals is number of fields
    // minus
    //      the number of chromosomes/dependent variables)
    // else
    //      use the number of terminals specified in the parameter file

    terminalFilename =
        state.parameters.getStringWithDefault(
            base.push(P_TERMINALFILENAME), def.push(P_TERMINALFILENAME), "");
    testingTerminalFilename =
        state.parameters.getStringWithDefault(
            base.push(P_TESTINGTERMINALFILENAME), def.push(P_TESTINGTERMINALFILENAME), "");
    String terminalSymbolsfromFile[] = null;
    CsvReader terminalFileCSV = null;
    CsvReader testingTerminalFileCSV = null;
    // Are we processing raw time series data?
    boolean timeseriesWithRawDataValues =
        species.problemType == GEPSpecies.PT_TIMESERIES && species.timeseriesEmbeddingDimension > 0;
    if (!terminalFilename.equals("")) {
      String defaultTerminalFileSeparator = ","; // default field separator is comma
      try {
        // allow for gzip files .... end with .gz or .gzip\
        if (terminalFilename.endsWith(".gz") || terminalFilename.endsWith(".gzip")) {
          terminalFileCSV =
              new CsvReader(
                  (InputStream) (new GZIPInputStream(new FileInputStream(terminalFilename))),
                  Charset.forName("ISO-8859-1"));
          // set terminal file name to be the one with gzip or gz removed from the end
          if (terminalFilename.endsWith(".gz"))
            terminalFilename = terminalFilename.substring(0, terminalFilename.length() - 3);
          else terminalFilename = terminalFilename.substring(0, terminalFilename.length() - 5);
        } else terminalFileCSV = new CsvReader(terminalFilename);
      } catch (FileNotFoundException e) {
        state.output.fatal(
            "The file with terminal definitions and/or values ("
                + terminalFilename
                + ") could not be found",
            base.push(P_TERMINALFILENAME),
            def.push(P_TERMINALFILENAME));
      } catch (IOException e) {
        state.output.fatal(
            "The file with terminal definitions and/or values ("
                + terminalFilename
                + ") could not be found or the expected GZIP file could nor be opened",
            base.push(P_TERMINALFILENAME),
            def.push(P_TERMINALFILENAME));
      }
      // if filename has extension .dat it is space delimited, if .csv (or anything else
      // for that matter) it is comma delimited
      // (separator can still be changed with the terminalfileseparator parameter)
      if (terminalFilename.endsWith(".dat")) defaultTerminalFileSeparator = "space";
      // if using a file for the terminals and their values then check for a non-default separator
      String terminalFileSeparator =
          state.parameters.getStringWithDefault(
              base.push(P_TERMINALFILESEPARATOR),
              def.push(P_TERMINALFILESEPARATOR),
              defaultTerminalFileSeparator);
      if (terminalFileSeparator.toLowerCase().equals("comma")) terminalFileSeparator = ",";
      else if (terminalFileSeparator == "\\t" || terminalFileSeparator.toLowerCase().equals("tab"))
        terminalFileSeparator = "\t";
      else if (terminalFileSeparator == "space") terminalFileSeparator = " ";
      terminalFileCSV.setDelimiter(terminalFileSeparator.charAt(0));
      // let's check for a testing data file at this time as well .. if no file for
      // names and training data no need to worry about this one.
      if (!testingTerminalFilename.equals("")) {
        try {
          // allow for gzip files .... end with .gz or .gzip\
          if (testingTerminalFilename.endsWith(".gz") || testingTerminalFilename.endsWith(".gzip"))
            testingTerminalFileCSV =
                new CsvReader(
                    (InputStream)
                        (new GZIPInputStream(new FileInputStream(testingTerminalFilename))),
                    Charset.forName("ISO-8859-1"));
          else testingTerminalFileCSV = new CsvReader(testingTerminalFilename);
          testingTerminalFileCSV.setDelimiter(terminalFileSeparator.charAt(0));
        } catch (FileNotFoundException e) {
          state.output.fatal(
              "The file with testing data values ("
                  + testingTerminalFilename
                  + ") could not be found",
              base.push(P_TERMINALFILENAME),
              def.push(P_TERMINALFILENAME));
        } catch (IOException e) {
          state.output.fatal(
              "The file with testing data values ("
                  + terminalFilename
                  + ") could not be found or the expected GZIP file could nor be opened",
              base.push(P_TERMINALFILENAME),
              def.push(P_TERMINALFILENAME));
        }
      }
    }

    if (timeseriesWithRawDataValues) numberOfTerminals = species.timeseriesEmbeddingDimension;
    else if (terminalFileCSV != null) {
      // get the terminal symbols for the independent and dependent variables
      try {
        terminalFileCSV.readHeaders();
        terminalSymbolsfromFile = terminalFileCSV.getHeaders();
      } catch (IOException e) {
        state.output.fatal(
            "The file with variable (terminal) definitions and values ("
                + terminalFilename
                + ") failed to read the headers"
                + e,
            base.push(P_TERMINALFILENAME),
            def.push(P_TERMINALFILENAME));
      }
      // 1 less for each dependent variable (number of chromosomes) at the end
      numberOfTerminals = terminalSymbolsfromFile.length - species.numberOfChromosomes;
      if (numberOfTerminals < 1)
        state.output.fatal(
            "The file with terminal definitions and data values ("
                + terminalFilename
                + ") has no independent variables specified in record 1",
            base.push(P_TERMINALFILENAME),
            def.push(P_TERMINALFILENAME));
      // if using a file for the terminals and their values then check for a non-default separator
    } else {
      numberOfTerminals =
          state.parameters.getInt(base.push(P_TERMINALSIZE), def.push(P_TERMINALSIZE), 1);
    }
    numberOfSymbols += numberOfTerminals;

    if (numberOfSymbols < 1)
      state.output.error(
          "The GEPSymbolSet \"" + name + "\" have at least 1 terminal symbol defined.",
          base.push(P_TERMINALSIZE),
          def.push(P_TERMINALSIZE));

    // add a special Symbol for constants if we are using them ... it will be added to the
    // end of the array of symbols!
    if (species.useConstants) {
      numberOfTerminals++; // special constant terminal
      numberOfSymbols++;
    }

    symbols = new GEPSymbol[numberOfSymbols];

    int numberOfSymbolsWithoutConstantSymbol = numberOfSymbols;
    if (species.useConstants) // add the constant terminal symbol to the end
    {
      symbols[numberOfSymbols - 1] = (GEPSymbol) (new GEPConstantTerminalSymbol());
      symbols[numberOfSymbols - 1].id = numberOfSymbols - 1;
      numberOfSymbolsWithoutConstantSymbol--;
    }

    Parameter pTerminal = base.push(P_TERMINAL);
    Parameter pdefTerminal = def.push(P_TERMINAL);
    Parameter pFunction = base.push(P_FUNCTION);
    Parameter pdefFunction = def.push(P_FUNCTION);

    // create hashtable of names of terminals and hash table with names of functions
    // so we can easily check that they are not duplicates
    Hashtable functionHT = new Hashtable();
    Hashtable terminalHT = new Hashtable();

    //      process the functions
    for (int x = 0; x < numberOfFunctions; x++) {
      Parameter pp = pFunction.push("" + x);
      Parameter ppdef = pdefFunction.push("" + x);
      String function = state.parameters.getStringWithDefault(pp, ppdef, "");
      if (function.equals("")) // no name for the function
      state.output.fatal("Invalid function specifier: '" + function + "'", pp, ppdef);
      // make sure not specifying the same function more than once
      if (functionHT.get(function) != null)
        state.output.fatal(
            "Function '" + function + "' was specified more than once in list of function symbols");
      else functionHT.put(function, function);
      GEPFunctionSymbol fs = null;
      try {
        Class classDefinition = Class.forName(LOCATION_OF_FUNCTION_CLASSES + "." + function);
        fs = (GEPFunctionSymbol) classDefinition.newInstance();
      } catch (InstantiationException e) {
        state.output.fatal(
            "Unable to create GEPFunctionSymbol class for function '" + function + "'. " + e);
      } catch (IllegalAccessException e) {
        state.output.fatal(
            "Unable to create GEPFunctionSymbol class for function '" + function + "' " + e);
      } catch (ClassNotFoundException e) {
        state.output.fatal(
            "Unable to create GEPFunctionSymbol class for function '" + function + "' " + e);
      }

      // if using a logical function must be a logical problem
      if (fs.isLogicalFunction() && (species.problemType != GEPSpecies.PT_LOGICAL))
        state.output.fatal(
            "Can only use logical functions with a logical problem type. Function "
                + function
                + " is  a logical function.",
            pp,
            ppdef);
      // if using a numerical function must be an non logical problem
      if (!fs.isLogicalFunction() && (species.problemType == GEPSpecies.PT_LOGICAL))
        state.output.fatal(
            "Can only use logical functions with a non logical problem type. Function "
                + function
                + " is a numerical function.",
            pp,
            ppdef);

      symbols[x] = (GEPSymbol) fs;
      // symbols[x].setup(state, base);
      if (fs.arity < 1) state.output.fatal("Arity must be > 0 for a GEPTerminalSymbol)", pp, ppdef);
      symbols[x].id = x;
      int weight =
          state.parameters.getInt(pp.push(P_FUNCTIONWEIGHT), ppdef.push(P_FUNCTIONWEIGHT), 1);
      if (weight < 1) {
        state.output.warning(
            "Weight for GEP Function must be > 0; defaulting to 1)",
            pp.push(P_FUNCTIONWEIGHT),
            ppdef.push(P_FUNCTIONWEIGHT));
        weight = 1;
      }
      symbols[x].weight = weight;
      if (symbols[x].arity > maxArity) maxArity = symbols[x].arity;
    }

    // process the terminals  ... defined by default for timeseries data, in the
    // CSV file if specified and not timeseries, or in the params file if neither of those.
    for (int x = numberOfFunctions;
        x < numberOfSymbolsWithoutConstantSymbol;
        x++) { // load the terminal symbols
      int index = x - numberOfFunctions;
      String terminal = "";
      if (timeseriesWithRawDataValues) {
        // terminals get default names v0, v1, v2, v3, ... vn-1
        terminal = "v" + index;
      } else if (terminalFileCSV == null) // terminals defined in param file
      {
        Parameter pp = pTerminal.push("" + index);
        Parameter ppdef = pdefTerminal.push("" + index);
        terminal = state.parameters.getStringWithDefault(pp, ppdef, "");
      } else { // terminals defined in CSV file
        terminal = terminalSymbolsfromFile[index];
      }
      if (terminal.equals("")) // no name for the terminal
      state.output.fatal("Invalid terminal specifier: '" + terminal + "' for terminal # " + index);
      // make sure not specifying the same function more than once
      if (terminalHT.get(terminal) != null)
        state.output.fatal(
            "Terminal symbol (indep var) '"
                + terminal
                + "' was specified more than once in list of terminal symbols (independent variables)");
      else terminalHT.put(terminal, terminal);
      GEPTerminalSymbol ts = new GEPTerminalSymbol(terminal, this);
      symbols[x] = (GEPSymbol) ts;
      // symbols[x].setup(state, base);
      if (ts.arity != 0) // cannot happen
      state.output.fatal("Arity must be exactly 0 for a GEPTerminalSymbol)");
      symbols[x].id = x;
      symbols[x].weight = 1; // all Terminal symbols have weight of 1
    }

    // must be at least 1 Terminal symbol in the SymbolSet.
    // If not then the user didn't specify the terminals in the param file or in the data file
    if (numberOfTerminals < 1)
      state.output.fatal(
          "Must be at least one Terminal Symbol in the set of GEPSymbols\n"
              + "Either did not specify the terminal symbols in the param file or\n"
              + "did not specify the appropriate data file with the terminals specified in the first line.");

    // collect the id's (indices) of the terminal and function symbols that
    // are in the set of symbols
    terminals = new int[numberOfTerminals];
    int terminalNum = 0;
    functions = new int[numberOfFunctions];
    int functionNum = 0;
    for (int x = 0; x < numberOfSymbols; x++) {
      if (symbols[x] instanceof GEPConstantTerminalSymbol) terminals[terminalNum++] = x;
      else if (symbols[x] instanceof GEPTerminalSymbol) terminals[terminalNum++] = x;
      else if (symbols[x] instanceof GEPFunctionSymbol) functions[functionNum++] = x;
    }

    // collect the weights for symbols and terminals and normalize and cumulate them.
    // Then we can use these arrays to pick appropriate symbols or terminals according to
    // their weights ... using the RandomChooser.PickFromDistribution
    cumulativeNormalizedSymbolWeights = new float[numberOfSymbols];
    cumulativeNormalizedTerminalWeights = new float[numberOfTerminals];
    cumulativeNormalizedFunctionWeights = new float[numberOfFunctions];
    int j = 0, k = 0;
    for (int i = 0; i < numberOfSymbols; i++) {
      float weight = (float) (symbols[i].weight);
      cumulativeNormalizedSymbolWeights[i] = weight;
      if (symbols[i] instanceof GEPTerminalSymbol
          || symbols[i] instanceof GEPConstantTerminalSymbol)
        cumulativeNormalizedTerminalWeights[j++] = weight;
      if (symbols[i] instanceof GEPFunctionSymbol)
        cumulativeNormalizedFunctionWeights[k++] = weight;
    }
    RandomChoice.organizeDistribution(cumulativeNormalizedSymbolWeights);
    RandomChoice.organizeDistribution(cumulativeNormalizedTerminalWeights);
    RandomChoice.organizeDistribution(cumulativeNormalizedFunctionWeights);

    // use the 2/3 rule if fewer functions else the 1/2 rule (don't count the constant
    // terminal here)
    if (numberOfFunctions < (numberOfTerminals - (species.useConstants ? 1 : 0)))
      probabilityOfChoosingFunction = 2.0 / 3.0;
    else probabilityOfChoosingFunction = 0.5;

    // ... and finally get the training and testing data values for the terminals and dependent
    // variable
    // and put them into the Terminal instances (creating a 'special' Terminal Symbol to
    // hold the dependent variable training and testing values)

    // If this is a time series problem AND we are using the raw time series data then
    // we named the terminals v1, v2, ..., nn where n is the number of independent
    // variables as specified in the embedding dimension (which) was used to
    // determine the number of terminals. But we have to process the time series data
    // to get the values for each terminal ... get the raw data from the CSV file
    // if specified or from the user program ... then process it into rows of data
    // representing the independent variables and the dependent variable.
    //
    //   timeseries-delay -- if 1 uses each time series value, if 2 uses every other one, etc.
    //   timeseries-embeddingdimension -- determines the number of timeseries points to use
    //        as independent variables when transforming the set of time series data. Another
    //        data point is used as the dependent variable value. So the time series 'raw' data
    //        consisting of a list of single values is processed by splitting the data into
    //        groups (rows) of size embeddingdimension+1. From the end of the time series data
    //        embeddingdimension+1 values are chosen (if delay is 1 all values are chosen, if
    //        2 every other one is chosen). The last value is the independent variable value.
    //        Then the next row is selected by moving 'delay'
    //        values from the end and chosing embeddingdimension+1 values. This is repeated
    //        until no more sets of size embeddingdimension+1 can be chosen. If this produces
    //        n sets of data then testingprediction of them are used for testing and
    //        (n - testingpredictions) are used for training.
    //
    //        So if we had the data:
    //        1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
    //        and delay was 1 and embeddingdimension was 4 then we'd process the set into
    //        the following 17 data sets. If testingpredictions was 6 then the 1st 11
    //        would be used for training and the last 6 for testing
    //        iv1 iv2 iv3 iv4  dv
    //          1   2   3   4   5
    //          2   3   4   5   6
    //          3   4   5   6   7
    //              . . .
    //         14  15  16  17  18
    //         15  16  17  18  19
    //         16  17  18  19  20
    //         17  18  19  20  21
    //        If delay was 2 then 7 sets would be formed as:
    //        iv1 iv2 iv3 iv4  dv
    //          1   3   5   7   9
    //          3   5   7   9  11
    //              . . .
    //          9  11  13  15  17
    //         11  13  15  17  19
    //         13  15  17  19  21
    //   timeseries-testingpredictions -- specifies the number of sets of data to devote to testing
    if (timeseriesWithRawDataValues) {
      GEPDependentVariable.symbol = "dependentVariable";
      double rawTimeSeriesValues[] = null;
      if (terminalFileCSV == null)
        rawTimeSeriesValues = ((GEPProblem) state.evaluator.p_problem).getTimeSeriesDataValues();
      else rawTimeSeriesValues = getRawTimeSeriesValuesFromCSVfile(state, terminalFileCSV);
      if (rawTimeSeriesValues == null)
        state.output.fatal("Unable to get time series data values from User Program or CSV file");
      Vector values[] = processRawTimeSeriesValues(state, species, rawTimeSeriesValues);
      // have an array of vectors; 1 vector for each indep variable and the dep variable(s)
      for (int i = 0; i < values.length; i++) {
        // get the values for training ... and testing (specified by timeseriesTestingPredictions)
        int sizeOfTrainingData = values[i].size() - species.timeseriesTestingPredictions;
        double v[] = new double[sizeOfTrainingData];
        double testingV[] = new double[species.timeseriesTestingPredictions];
        for (int m = 0; m < v.length; m++) v[m] = ((Double) values[i].elementAt(m)).doubleValue();
        for (int n = 0; n < testingV.length; n++)
          testingV[n] = ((Double) values[i].elementAt(n + sizeOfTrainingData)).doubleValue();
        int depVarIndex = i - values.length + species.numberOfChromosomes;
        if (depVarIndex >= 0) // last column(s) in file is(are) the dependent variable(s)
        {
          GEPDependentVariable.trainingData.setValues(v, depVarIndex);
          GEPDependentVariable.testingData.setValues(testingV, depVarIndex);
        } else {
          ((GEPTerminalSymbol) symbols[numberOfFunctions + i]).setTrainingValues(v);
          ((GEPTerminalSymbol) symbols[numberOfFunctions + i]).setTestingValues(testingV);
        }
      }
    }
    // else If there is a file with the terminals and dep variable(s) use this else ask for
    // the values from the User Program (problem).
    else if (terminalFileCSV != null) // terminals defined in CSV file
    {
      GEPDependentVariable.symbol = terminalSymbolsfromFile[terminalSymbolsfromFile.length - 1];
      // get all the values into an array of vectors (each vector holds the values for a
      // single terminal (dep or indep variable)
      Vector values[] = new Vector[terminalSymbolsfromFile.length];
      for (int i = 0; i < terminalSymbolsfromFile.length; i++) values[i] = new Vector();
      try {
        while (terminalFileCSV.readRecord()) {
          for (int i = 0; i < terminalSymbolsfromFile.length; i++)
            values[i].add(terminalFileCSV.get(i));
        }
      } catch (IOException e) {
        state.output.fatal(
            "The file with terminal definitions/values failed when reading records. " + e);
      }

      for (int i = 0; i < terminalSymbolsfromFile.length; i++) {
        double v[] = new double[values[i].size()];
        for (int m = 0; m < v.length; m++)
          try {
            v[m] = Double.parseDouble((String) values[i].elementAt(m));
          } catch (Exception e) {
            state.output.fatal(
                "Failed trying to read a training data set value. The field is supposed to be a number but was the string '"
                    + (String) values[i].elementAt(m)
                    + "'.\n"
                    + e);
          }
        int jj = terminalSymbolsfromFile.length - species.numberOfChromosomes;
        if (i >= jj) // last column(s) in file is(are) the dependent variable(s)
        GEPDependentVariable.trainingData.setValues(v, i - jj);
        else ((GEPTerminalSymbol) symbols[numberOfFunctions + i]).setTrainingValues(v);
      }
      // get the testing data as well if a file was specified
      if (testingTerminalFileCSV != null) // testing data defined in CSV file
      {
        // get all the values into an array of vectors (each vector holds the values for a
        // single terminal (dep or indep variable)
        Vector testingValues[] = new Vector[terminalSymbolsfromFile.length];
        for (int i = 0; i < terminalSymbolsfromFile.length; i++) testingValues[i] = new Vector();
        try {
          while (testingTerminalFileCSV.readRecord()) {
            for (int i = 0; i < terminalSymbolsfromFile.length; i++)
              testingValues[i].add(testingTerminalFileCSV.get(i));
          }
        } catch (IOException e) {
          state.output.fatal(
              "The file with testing data values failed when reading records. "
                  + "\nMake sure the file has the same column separators as the testing data file."
                  + "\nAlso check that it has the same as the number of columns as the testing file"
                  + e);
        }

        for (int i = 0; i < terminalSymbolsfromFile.length; i++) {
          double v[] = new double[testingValues[i].size()];
          for (int m = 0; m < v.length; m++)
            try {
              v[m] = Double.parseDouble((String) testingValues[i].elementAt(m));
            } catch (Exception e) {
              state.output.fatal(
                  "Failed trying to read a testing data set value. The field is supposed to be a number but was the string '"
                      + (String) testingValues[i].elementAt(m)
                      + "'.\n"
                      + e);
            }
          int jj = terminalSymbolsfromFile.length - species.numberOfChromosomes;
          if (i >= jj) // last column(s) in file is(are) the dependent variable(s)
          GEPDependentVariable.testingData.setValues(v, i - jj);
          else ((GEPTerminalSymbol) symbols[numberOfFunctions + i]).setTestingValues(v);
        }
      }
    }
    // else terminals were defined in the param file and no CSV file
    // defined so .... ask User Problem for the values, training and testing (if there are any)
    else {
      GEPDependentVariable.symbol = "dependentVariable";
      GEPProblem prob = (GEPProblem) state.evaluator.p_problem;
      double vals[] = null;
      for (int i = numberOfFunctions; i < numberOfSymbolsWithoutConstantSymbol; i++) {
        GEPTerminalSymbol ts = (GEPTerminalSymbol) symbols[i];
        vals = prob.getDataValues(ts.symbol);
        if (vals == null)
          state.output.fatal(
              "Expecting user problem (GEPProblem/ProblemForm) to supply training data values for terminal symbol '"
                  + ts
                  + "'.");
        ts.setTrainingValues(vals);
        vals = prob.getTestingDataValues(ts.symbol);
        if (vals != null) // don't have to supply testing data
        ts.setTestingValues(vals);
      }
      // if just one dep var then ask user by requesting with getdataValues("dependentVariable")
      // and if more than one dep var (more than 1 chromosome) then ask for dep variables
      // with getDataValues("dependentVariable0"), getDataValues("dependentVariable1"), ...
      for (int i = 0; i < species.numberOfChromosomes; i++) {
        String depVarSym = GEPDependentVariable.symbol;
        if (species.numberOfChromosomes > 1) depVarSym = depVarSym + i;
        vals = prob.getDataValues(depVarSym);
        if (vals == null)
          state.output.fatal(
              "Expecting user problem (GEPProblem/ProblemForm) to supply training data values for dependent variable '"
                  + depVarSym
                  + "'.");
        GEPDependentVariable.trainingData.setValues(vals, i);
        vals = prob.getTestingDataValues(depVarSym);
        if (vals != null) // don't have to supply testing data
        GEPDependentVariable.testingData.setValues(vals, i);
      }
    }

    // Some checking of data values to ensure they meet the requirements for the various problem
    // types.
    // For all problem types need to make sure all indep vars and the dep var have the same number
    // of values!
    int numValues = GEPDependentVariable.trainingData.values[0].length;
    for (int i = numberOfFunctions; i < numberOfSymbolsWithoutConstantSymbol; i++)
      if (((GEPTerminalSymbol) symbols[i]).trainingValues.length != numValues)
        state.output.fatal(
            "Must have same number of values for all independent variables and the dependent variable."
                + "/nNumber of values for Dependent Variable is: "
                + numValues
                + "/nNumber of values for Independent Variable '"
                + symbols[i].symbol
                + "' is: "
                + ((GEPTerminalSymbol) symbols[i]).trainingValues.length);
    // For Classification and logical problems all dependent variable values must be either 0 or 1
    if (species.problemType == GEPSpecies.PT_CLASSIFICATION
        || species.problemType == GEPSpecies.PT_LOGICAL) {
      double dvVals[] = GEPDependentVariable.trainingData.values[0];
      for (int i = 0; i < numValues; i++)
        if (dvVals[i] != 0.0 && dvVals[i] != 1.0)
          state.output.fatal(
              "For classification/logical problems all dependent variable values must be either 1 or 0.\nFound value "
                  + dvVals[i]
                  + " at index "
                  + i
                  + "in the values.");
    }
    // For Logical problems all independent variable values must be 0 or 1
    if (species.problemType == GEPSpecies.PT_LOGICAL) { // for each indep variable symbol
      for (int i = numberOfFunctions; i < numberOfSymbolsWithoutConstantSymbol; i++) {
        double ivVals[] = ((GEPTerminalSymbol) symbols[i]).trainingValues;
        for (int m = 0; m < numValues; m++)
          if (ivVals[m] != 0.0 && ivVals[m] != 1.0)
            state.output.fatal(
                "For logical problems all independent variable values must be either 1 or 0.\nFound value "
                    + ivVals[m]
                    + " at index '"
                    + m
                    + "' in the variable '"
                    + ((GEPTerminalSymbol) symbols[i]).symbol
                    + "'.");
      }
    }
    state.output.exitIfErrors();
  }
예제 #3
0
  private static List<String[]> readRecords(
      final Hashtable<String, String> global, final CsvReader reader) throws IOException {
    final List<String[]> transactions = new ArrayList<String[]>();
    reader.setDelimiter(';');
    reader.setUseComments(false);
    reader.setTextQualifier('\'');

    int state = 0;

    while (reader.readRecord()) {
      final String[] record = reader.getValues();
      final String first = record[0];
      switch (state) {
        case 0:
          if (first.startsWith("#Klient")) {
            if (reader.readRecord()) {
              global.put("name", reader.getValues()[0]);
              state = 1;
            }
          }
          break;
        case 1:
          if (first.startsWith("#Za okres:")) {
            if (reader.readRecord()) {
              global.put("start_date", reader.getValues()[0]);
              global.put("end_date", reader.getValues()[1]);
              state = 2;
            }
          }
          break;
        case 2:
          if (first.startsWith("#Waluta")) {
            if (reader.readRecord()) {
              global.put("currency", reader.getValues()[0]);
              state = 3;
            }
          }
          break;
        case 3:
          if (first.startsWith("#Numer rachunku")) {
            if (reader.readRecord()) {
              global.put("account_number", reader.getValues()[0]);
              state = 4;
            }
          }
          break;
        case 4:
          if (first.startsWith("#Saldo pocz")) {
            global.put("start_balance", reader.getValues()[1]);
            state = 5;
          }
          break;
        case 5:
          if (first.startsWith("#Data operacji")) {
            state = 6;
          }
          break;
        case 6:
          if (record.length > 6 && record[6].startsWith("#Saldo ko")) {
            global.put("end_balance", record[4]);
            state = 7;
            continue;
          }
          if (record.length != 9) continue;
          transactions.add(record);
          break;
      }
    }
    return transactions;
  }
  public PaymentStatistic processKZ(File csvFile, LoggingHandler lh) throws IOException {

    CsvReader reader = null;
    ArrayList<PaymentRecord> recordList = new ArrayList();
    Double sum = 0.0;
    PaymentStatistic statisticAll = new PaymentStatistic();

    try {

      reader = new CsvReader(csvFile.getCanonicalPath());
      reader.setDelimiter(';');
      reader.readHeaders();

      // Headers ohne Spaces mittels trim
      String[] headers = new String[reader.getHeaderCount()];
      for (int i = 0; i < reader.getHeaderCount(); i++) {
        headers[i] = reader.getHeader(i).trim();
      }
      reader.setHeaders(headers);
      // http://www.csvreader.com/java_csv_samples.php
      // http://javacsv.sourceforge.net/

      PaymentRecord record = new PaymentRecord(reader);
      recordList.add(record);

      while (true) {
        try {

          PaymentRecord newRecord = new PaymentRecord(reader);

          if (newRecord.getKontoNr() == record.getKontoNr()) {
            // nicht verarbeiten, gleicher Kunde
            recordList.add(newRecord);
            sum = sum + newRecord.getBetragNr();
            record = newRecord;
            continue;
          }

          // unterschiedliche Kundennummer, alte Belege verarbeiten
          if (sum == 0.0) {
            PaymentStatistic statistic = processKZPaymentRecord(recordList, lh);
            statisticAll.addStatistic(statistic);
          }
          // System.out.println("Kunde " + record.getKontoStr() + "OP: " + sum);
          recordList = new ArrayList();
          sum = 0.0;
          recordList.add(newRecord);
          sum = sum + newRecord.getBetragNr();
          record = newRecord;
          continue;

        } catch (Throwable th) {
          // kein neuer record mehr, Liste verarbeiten
          if (sum == 0.0) {
            PaymentStatistic statistic = processKZPaymentRecord(recordList, lh);
            statisticAll.addStatistic(statistic);
          }
        }
        break;
      }

      reader.close();

    } catch (Throwable th) {
      lh.getLogger(Globals.LOGGINGKZ).log(Level.SEVERE, null, th);
      return statisticAll;
    }
    return statisticAll;
  }